MC@NLO event generation by reweighting unweighted Born events

Abstract

We propose a computational strategy for NLO+PS simulations in the MC@NLO framework that starts from Born-accurate (LO) events and reweights them to the full MC@NLO S-event weight, while generating H-events separately. We validate the approach on two representative LHC processes and compare to direct NLO event generation for both standard MC@NLO and MC@NLO-Delta matching. Employing large folding values in the radiative variables stabilizes the S-event integral, reduces weight variance, and significantly lowers the fraction of negative weights compared to S-event generation without folding. At fixed precision, this pipeline has comparable wall-clock times relative to standard S-event generation and unweighting, with room for further optimisation.

Figures (2)

• Figure 1: Differential distributions for the $\mathbb{S}$-event contribution to the process $pp \to t\bar{t}$ comparing reweighted LO samples and directly generated NLO S-events. The top row shows results with standard MC@NLO matching: the $t\bar{t}$ invariant mass $m_{t\bar{t}}$ (left) and the transverse momentum of the $t\bar{t}$ system $p_{T}^{t\bar{t}}$ (right). The bottom row shows the corresponding results for MC@NLO-$\Delta$: the top-quark rapidity $y_t$ (left) and $p_{T}^{t\bar{t}}$ (right). For each observable, we compare reweighted samples without folding and with high folding $(16,16,1)$ to directly generated NLO S-events without folding, with uncertainty bands indicating statistical errors.
• Figure 2: Differential distributions for the $\mathbb{S}$-event contribution to the process $pp \to e^{+}e^{-}j$ comparing reweighted LO samples and directly generated NLO S-events. The top row shows MC@NLO matching: the dilepton invariant mass $m_{e^{+}e^{-}}$ (left) and the pseudo-rapidity of the hardest jet $\eta_{j_1}$ (right). The bottom row shows the MC@NLO-$\Delta$ results: the angular separation between the dilepton system and the hardest jet, $\Delta R(e^{+}e^{-},j_1)$ (left), and the rapidity of the $e^{+}e^{-}$ pair (right). Reweighted samples without folding suffer from large statistical uncertainties due to the broad weight distribution, whereas applying high folding $(16,16,1)$ substantially reduces variance and leads to excellent agreement with the directly generated NLO S-event predictions.